fix(gateway): wire presence grace, request limits and voice sync (#3329)

fluxer_docs/src/content/docs/gateway/limits-and-rate-limits.md

@@ -48,7 +48,7 @@ Presence Update accepts five commands per WebSocket in a rolling 20-second windo
 
 Voice State Update processes the first two commands per session in a rolling one-second window immediately. Later updates enter a per-session queue that holds at most 64 commands and drains one command every 500 ms. A newer update replaces an older queued update for the same `guild_id` and `connection_id` pair, and a full queue discards its oldest entry before accepting the new one.
 
-Request Guild Members has one command-specific budget. A bot requesting a complete member list is limited to one accepted request per guild every 30 seconds, and a request inside that window produces [Rate Limited](/gateway/events/#rate-limited). The budget is keyed by the account and the guild together, so reconnecting does not reset it.
+Request Guild Members has three command-specific budgets. Each account may send 12 requests in a rolling 10-second window, and each guild accepts 40 requests in a rolling 10-second window across all accounts. A request over either of these budgets is discarded without a result and without a close. A bot requesting a complete member list is limited to one accepted request per guild every 30 seconds, and a request inside that window produces [Rate Limited](/gateway/events/#rate-limited). The budget is keyed by the account and the guild together, so reconnecting does not reset it.
 
 ## Bounded commands
 

fluxer_gateway/src/gateway/gateway_cluster_membership.erl

@@ -140,8 +140,10 @@ handle_cast(_Msg, State) ->
 handle_info({cluster_peers_changed, Peers}, State) ->
     {noreply, apply_discovered(normalize_nodes(Peers), State)};
 handle_info({nodeup, Node}, State) when is_atom(Node) ->
+    ok = presence_cache:handle_nodeup(Node),
     {noreply, maybe_add_member(Node, State)};
 handle_info({nodedown, Node}, State) when is_atom(Node) ->
+    ok = maybe_start_presence_grace(Node, State),
     {noreply, maybe_remove_member(Node, State)};
 handle_info(refresh_roles, State) ->
     State1 = reconcile_against_connected_nodes(State),
@@ -211,6 +213,13 @@ force_discovery_refresh_safely() ->
         exit:_ -> ok
     end.
 
+-spec maybe_start_presence_grace(node(), state()) -> ok.
+maybe_start_presence_grace(Node, #{members := Members}) ->
+    case lists:member(Node, Members) andalso Node =/= node() of
+        true -> presence_cache:handle_nodedown(Node);
+        false -> ok
+    end.
+
 -spec maybe_remove_member(node(), state()) -> state().
 maybe_remove_member(Node, #{members := Members} = State) ->
     case lists:member(Node, Members) andalso Node =/= node() of

fluxer_gateway/src/guild/guild.erl

@@ -516,6 +516,12 @@ terminate(Reason, State) when is_map(State) ->
         end,
         "ets_cleanup"
     ),
+    safe_cleanup(
+        fun() ->
+            cleanup_passive_sync_registry(State)
+        end,
+        "passive_sync_cleanup"
+    ),
     safe_cleanup(
         fun() ->
             cleanup_voice_server(State)
@@ -578,6 +584,15 @@ cleanup_per_guild_ets(State) ->
     safe_delete_ets(maps:get(viewable_channels_cache, State, undefined)),
     ok.
 
+-spec cleanup_passive_sync_registry(guild_state()) -> ok.
+cleanup_passive_sync_registry(#{id := GuildId} = State) when is_integer(GuildId) ->
+    maps:foreach(
+        fun(SessionId, _Session) -> passive_sync_registry:delete(SessionId, GuildId) end,
+        maps:get(sessions, State, #{})
+    );
+cleanup_passive_sync_registry(_State) ->
+    ok.
+
 -spec cleanup_voice_server(guild_state()) -> ok.
 cleanup_voice_server(State) ->
     case maps:get(voice_server_pid, State, undefined) of

fluxer_gateway/src/guild/guild_request_members.erl

@@ -180,9 +180,16 @@ normalize_nonce(_) ->
 
 -spec process_request(map(), pid(), session_state()) -> ok | {error, atom()}.
 process_request(Request, SocketPid, SessionState) ->
+    UserId = snowflake_id:parse_optional(maps:get(user_id, SessionState)),
+    case guild_request_members_filter:check_request_rate_limit(UserId) of
+        ok -> process_admitted_request(Request, UserId, SocketPid, SessionState);
+        {error, Reason} -> {error, Reason}
+    end.
+
+-spec process_admitted_request(map(), integer() | undefined, pid(), session_state()) ->
+    ok | {error, atom()}.
+process_admitted_request(Request, UserId, SocketPid, SessionState) ->
     #{guild_ids := ReqGuildIds, query := Query, limit := Limit, user_ids := UserIds} = Request,
-    UserIdBin = maps:get(user_id, SessionState),
-    UserId = snowflake_id:parse_optional(UserIdBin),
     IsBot = truthy(maps:get(bot, SessionState, false)),
     case guild_request_members_filter:enforce_single_guild_for_bots(IsBot, ReqGuildIds) of
         ok ->
@@ -248,6 +255,22 @@ dispatch_guild_requests(GIds, UID, Bot, Q, L, UI, Req, SP, SS) ->
     session_state()
 ) -> ok | {error, atom()}.
 process_single_guild(GId, UID, Bot, Q, L, UI, Req, _SP, SS) ->
+    case guild_request_members_filter:check_guild_request_rate_limit(GId) of
+        ok -> process_admitted_guild(GId, UID, Bot, Q, L, UI, Req, SS);
+        {error, Reason} -> {error, Reason}
+    end.
+
+-spec process_admitted_guild(
+    integer(),
+    integer() | undefined,
+    boolean(),
+    binary(),
+    non_neg_integer(),
+    [integer()],
+    map(),
+    session_state()
+) -> ok | {error, atom()}.
+process_admitted_guild(GId, UID, Bot, Q, L, UI, Req, SS) ->
     IsFull = guild_request_members_filter:is_full_member_list(Q, L, UI),
     case guild_request_members_filter:check_full_list_bot_rate_limit(IsFull, Bot, UID, GId) of
         ok ->
@@ -320,6 +343,22 @@ try_guild_or_keep_processed(GId, UID, Bot, Q, L, UI, Req, SS, Acc) ->
     session_state()
 ) -> boolean().
 try_guild(GId, UID, Bot, Q, L, UI, Req, SS) ->
+    case guild_request_members_filter:check_guild_request_rate_limit(GId) of
+        ok -> try_admitted_guild(GId, UID, Bot, Q, L, UI, Req, SS);
+        {error, _Reason} -> false
+    end.
+
+-spec try_admitted_guild(
+    integer(),
+    integer() | undefined,
+    boolean(),
+    binary(),
+    non_neg_integer(),
+    [integer()],
+    map(),
+    session_state()
+) -> boolean().
+try_admitted_guild(GId, UID, Bot, Q, L, UI, Req, SS) ->
     IsFull = guild_request_members_filter:is_full_member_list(Q, L, UI),
     case guild_request_members_filter:check_full_list_bot_rate_limit(IsFull, Bot, UID, GId) of
         ok ->

fluxer_gateway/src/guild/guild_request_members_filter.erl

@@ -16,6 +16,10 @@
 
 -export_type([session_state/0]).
 
+-ifdef(TEST).
+-export([prune_expired_entries/2]).
+-endif.
+
 -define(REQUEST_MEMBERS_RATE_LIMIT_TABLE, guild_request_members_rate_limit).
 -define(REQUEST_MEMBERS_RATE_LIMIT_WINDOW_MS, 10000).
 -define(REQUEST_MEMBERS_RATE_LIMIT_MAX_EVENTS, 12).
@@ -24,6 +28,7 @@
 -define(REQUEST_MEMBERS_GUILD_RATE_LIMIT_MAX_EVENTS, 40).
 -define(FULL_LIST_BOT_RATE_LIMIT_TABLE, guild_request_members_bot_full_list_rate_limit).
 -define(FULL_LIST_BOT_RATE_LIMIT_WINDOW_MS, 30000).
+-define(RATE_LIMIT_CLEANUP_INTERVAL_MS, 60000).
 
 -type session_state() :: map().
 
@@ -106,7 +111,7 @@ check_full_list_bot_rate_limit(true, true, _UserId, _GuildId) ->
 -spec check_bot_rate_limit_window(integer(), integer()) ->
     ok | {rate_limited, non_neg_integer()}.
 check_bot_rate_limit_window(UserId, GuildId) ->
-    ensure_ets_table(?FULL_LIST_BOT_RATE_LIMIT_TABLE),
+    ensure_ets_table(?FULL_LIST_BOT_RATE_LIMIT_TABLE, ?FULL_LIST_BOT_RATE_LIMIT_WINDOW_MS),
     Now = erlang:system_time(millisecond),
     Key = {UserId, GuildId},
     Window = ?FULL_LIST_BOT_RATE_LIMIT_WINDOW_MS,
@@ -120,25 +125,40 @@ check_bot_rate_limit_window(UserId, GuildId) ->
 
 -spec check_request_rate_limit(integer() | undefined) -> ok | {error, atom()}.
 check_request_rate_limit(UserId) when is_integer(UserId), UserId > 0 ->
-    ensure_ets_table(?REQUEST_MEMBERS_RATE_LIMIT_TABLE),
-    check_sliding_window(
-        ?REQUEST_MEMBERS_RATE_LIMIT_TABLE,
-        UserId,
-        ?REQUEST_MEMBERS_RATE_LIMIT_WINDOW_MS,
-        ?REQUEST_MEMBERS_RATE_LIMIT_MAX_EVENTS
-    );
+    case gateway_handler_rate_limit:rate_limits_disabled() of
+        true ->
+            ok;
+        false ->
+            ensure_ets_table(
+                ?REQUEST_MEMBERS_RATE_LIMIT_TABLE, ?REQUEST_MEMBERS_RATE_LIMIT_WINDOW_MS
+            ),
+            check_sliding_window(
+                ?REQUEST_MEMBERS_RATE_LIMIT_TABLE,
+                UserId,
+                ?REQUEST_MEMBERS_RATE_LIMIT_WINDOW_MS,
+                ?REQUEST_MEMBERS_RATE_LIMIT_MAX_EVENTS
+            )
+    end;
 check_request_rate_limit(_) ->
     {error, invalid_session}.
 
 -spec check_guild_request_rate_limit(integer()) -> ok | {error, atom()}.
 check_guild_request_rate_limit(GuildId) when is_integer(GuildId), GuildId > 0 ->
-    ensure_ets_table(?REQUEST_MEMBERS_GUILD_RATE_LIMIT_TABLE),
-    check_sliding_window(
-        ?REQUEST_MEMBERS_GUILD_RATE_LIMIT_TABLE,
-        GuildId,
-        ?REQUEST_MEMBERS_GUILD_RATE_LIMIT_WINDOW_MS,
-        ?REQUEST_MEMBERS_GUILD_RATE_LIMIT_MAX_EVENTS
-    );
+    case gateway_handler_rate_limit:rate_limits_disabled() of
+        true ->
+            ok;
+        false ->
+            ensure_ets_table(
+                ?REQUEST_MEMBERS_GUILD_RATE_LIMIT_TABLE,
+                ?REQUEST_MEMBERS_GUILD_RATE_LIMIT_WINDOW_MS
+            ),
+            check_sliding_window(
+                ?REQUEST_MEMBERS_GUILD_RATE_LIMIT_TABLE,
+                GuildId,
+                ?REQUEST_MEMBERS_GUILD_RATE_LIMIT_WINDOW_MS,
+                ?REQUEST_MEMBERS_GUILD_RATE_LIMIT_MAX_EVENTS
+            )
+    end;
 check_guild_request_rate_limit(_) ->
     {error, invalid_guild_id}.
 
@@ -166,30 +186,56 @@ check_window_count(Table, Key, Now, Recent, _MaxEvents) ->
     ets:insert(Table, {Key, [Now | Recent]}),
     ok.
 
--spec ensure_ets_table(atom()) -> ok.
-ensure_ets_table(Name) ->
+-spec ensure_ets_table(atom(), pos_integer()) -> ok.
+ensure_ets_table(Name, WindowMs) ->
     case ets:whereis(Name) of
         undefined ->
-            create_ets_table(Name);
+            create_ets_table(Name, WindowMs);
         _ ->
             ok
     end.
 
--spec create_ets_table(atom()) -> ok.
-create_ets_table(Name) ->
+-spec create_ets_table(atom(), pos_integer()) -> ok.
+create_ets_table(Name, WindowMs) ->
     try
-        _ = ets:new(Name, [
-            named_table,
-            public,
-            set,
-            {read_concurrency, true},
-            {write_concurrency, true}
-        ]),
+        Tid = ets:new(
+            Name,
+            [
+                named_table,
+                public,
+                set,
+                {read_concurrency, true},
+                {write_concurrency, true}
+            ] ++ guild_ets_utils:heir_options()
+        ),
+        _ = spawn(fun() -> cleanup_loop(Tid, WindowMs) end),
         ok
     catch
         error:badarg -> ok
     end.
 
+-spec cleanup_loop(ets:table(), pos_integer()) -> ok.
+cleanup_loop(Table, WindowMs) ->
+    ok = gateway_retry_timer:wait(?RATE_LIMIT_CLEANUP_INTERVAL_MS),
+    case prune_expired_entries(Table, WindowMs) of
+        ok -> cleanup_loop(Table, WindowMs);
+        gone -> ok
+    end.
+
+-spec prune_expired_entries(ets:table(), pos_integer()) -> ok | gone.
+prune_expired_entries(Table, WindowMs) ->
+    Cutoff = erlang:system_time(millisecond) - WindowMs,
+    try
+        _ = ets:select_delete(Table, [
+            {{'_', []}, [], [true]},
+            {{'_', '$1'}, [{is_list, '$1'}, {'=<', {hd, '$1'}, Cutoff}], [true]},
+            {{'_', '$1'}, [{is_integer, '$1'}, {'=<', '$1', Cutoff}], [true]}
+        ]),
+        ok
+    catch
+        error:badarg -> gone
+    end.
+
 -spec dispatch_full_list_rate_limited(
     session_state(), integer(), map(), non_neg_integer()
 ) -> ok.

fluxer_gateway/src/guild/guild_state.erl

@@ -155,10 +155,7 @@ post_update_role(guild_role_delete, EventData, OldState, NewState) ->
     guild_state().
 resync_roles_after_permission_change(RoleIds, OldState, NewState) ->
     Recomputed = guild_state_roles:recompute_visibility_for_roles(RoleIds, OldState, NewState),
-    lists:foreach(
-        fun(RoleId) -> guild_voice_permission_sync:sync_users_with_role(RoleId, Recomputed) end,
-        RoleIds
-    ),
+    ok = guild_voice_permission_sync:sync_roles_after_change(RoleIds, OldState, Recomputed),
     Recomputed.
 
 -spec post_update_channel(event(), event_data(), guild_state(), guild_state()) -> guild_state().

fluxer_gateway/src/guild/passive_sync_registry.erl

@@ -7,8 +7,7 @@
     init/0,
     store/3,
     lookup/2,
-    delete/2,
-    delete_all_for_session/1
+    delete/2
 ]).
 
 -ifdef(TEST).
@@ -72,12 +71,6 @@ delete(SessionId, GuildId) ->
     ets:delete(?TABLE, Key),
     ok.
 
--spec delete_all_for_session(session_id()) -> ok.
-delete_all_for_session(SessionId) ->
-    ensure_table(),
-    ets:match_delete(?TABLE, {{SessionId, '_'}, '_'}),
-    ok.
-
 -spec ensure_table() -> ok.
 ensure_table() ->
     case ets:whereis(?TABLE) of
@@ -152,38 +145,6 @@ delete_removes_entry_test() ->
     ),
     cleanup_table().
 
-delete_all_for_session_removes_all_guilds_test() ->
-    cleanup_table(),
-    ok = init(),
-    SessionId = <<"session_1">>,
-    OtherSessionId = <<"session_2">>,
-    State1 = make_passive_state(<<"ch1">>, <<"msg1">>),
-    State2 = make_passive_state(<<"ch2">>, <<"msg2">>),
-    OtherState = make_passive_state(<<"ch3">>, <<"msg3">>),
-    ok = store(SessionId, 100, State1),
-    ok = store(SessionId, 200, State2),
-    ok = store(OtherSessionId, 100, OtherState),
-    ok = delete_all_for_session(SessionId),
-    Empty = empty_passive_state(),
-    ?assertEqual(Empty, lookup(SessionId, 100)),
-    ?assertEqual(Empty, lookup(SessionId, 200)),
-    ?assertEqual(OtherState, lookup(OtherSessionId, 100)),
-    cleanup_table().
-
-make_passive_state(Ch, Msg) ->
-    #{
-        previous_passive_updates => #{Ch => Msg},
-        previous_passive_channel_versions => #{},
-        previous_passive_voice_states => #{}
-    }.
-
-empty_passive_state() ->
-    #{
-        previous_passive_updates => #{},
-        previous_passive_channel_versions => #{},
-        previous_passive_voice_states => #{}
-    }.
-
 store_overwrites_existing_test() ->
     cleanup_table(),
     ok = init(),

fluxer_gateway/src/guild/voice/guild_voice_permission_sync.erl

@@ -6,8 +6,7 @@
 -export([
     sync_user_voice_permissions/2,
     sync_all_voice_permissions_for_channels/2,
-    sync_users_with_role/2,
-    maybe_sync_permissions_on_role_update/2,
+    sync_roles_after_change/3,
     maybe_sync_permissions_on_member_update/2
 ]).
 
@@ -96,14 +95,6 @@ sync_channel_voice_state_permissions(GuildId, VoiceState, State) ->
         UserId -> sync_voice_state_permissions(GuildId, UserId, VoiceState, State)
     end.
 
--spec maybe_sync_permissions_on_role_update(map(), guild_state()) -> ok.
-maybe_sync_permissions_on_role_update(RoleUpdate, State) ->
-    RoleId = snowflake_id:parse_optional(maps:get(<<"id">>, RoleUpdate, undefined)),
-    case RoleId of
-        undefined -> ok;
-        RoleIdInt -> sync_users_with_role(RoleIdInt, State)
-    end.
-
 -spec maybe_sync_permissions_on_member_update(map(), guild_state()) -> ok.
 maybe_sync_permissions_on_member_update(MemberUpdate, State) ->
     UserId = get_member_user_id(MemberUpdate),
@@ -241,54 +232,66 @@ enforce_voice_permissions_in_livekit(
             ok
     end.
 
--spec sync_users_with_role(integer(), guild_state()) -> ok.
-sync_users_with_role(RoleId, State) ->
-    VoiceStates = guild_voice_lifecycle:authoritative_voice_states(State),
+-spec sync_roles_after_change([integer()], guild_state(), guild_state()) -> ok.
+sync_roles_after_change([], _OldState, _State) ->
+    ok;
+sync_roles_after_change(RoleIds, OldState, State) ->
     case state_guild_id(State) of
         undefined ->
             ok;
         GuildId ->
-            do_sync_users_with_role(GuildId, RoleId, VoiceStates, State)
+            Affected = affected_role_users(GuildId, RoleIds, OldState, State),
+            sync_affected_voice_states(GuildId, Affected, State)
     end.
 
--spec do_sync_users_with_role(integer(), integer(), map(), guild_state()) -> ok.
-do_sync_users_with_role(GuildId, RoleId, VoiceStates, State) ->
-    RoleUsers = user_ids_with_role(RoleId, State),
+-spec sync_affected_voice_states(integer(), all | sets:set(user_id()), guild_state()) -> ok.
+sync_affected_voice_states(GuildId, Affected, State) ->
     maps:foreach(
         fun(_ConnId, VoiceState) ->
-            maybe_sync_role_voice_state(GuildId, RoleUsers, VoiceState, State)
+            maybe_sync_affected_voice_state(GuildId, Affected, VoiceState, State)
         end,
-        VoiceStates
-    ),
-    ok.
+        guild_voice_lifecycle:authoritative_voice_states(State)
+    ).
 
--spec maybe_sync_role_voice_state(integer(), sets:set(user_id()), voice_state(), guild_state()) ->
-    ok.
-maybe_sync_role_voice_state(GuildId, RoleUsers, VoiceState, State) ->
-    UserId = voice_state_utils:voice_state_user_id(VoiceState),
-    case UserId of
-        undefined ->
-            ok;
-        _ ->
-            sync_voice_state_permissions_if_role_matches(
-                GuildId, RoleUsers, UserId, VoiceState, State
+-spec affected_role_users(integer(), [integer()], guild_state(), guild_state()) ->
+    all | sets:set(user_id()).
+affected_role_users(GuildId, RoleIds, OldState, State) ->
+    case lists:member(GuildId, RoleIds) of
+        true ->
+            all;
+        false ->
+            OldIndex = guild_data_index:member_role_index(maps:get(data, OldState, #{})),
+            NewIndex = guild_data_index:member_role_index(maps:get(data, State, #{})),
+            sets:from_list(
+                lists:append([
+                    maps:keys(maps:get(RoleId, OldIndex, #{})) ++
+                        maps:keys(maps:get(RoleId, NewIndex, #{}))
+                 || RoleId <- RoleIds
+                ])
             )
     end.
 
--spec sync_voice_state_permissions_if_role_matches(
-    integer(), sets:set(user_id()), user_id(), voice_state(), guild_state()
+-spec maybe_sync_affected_voice_state(
+    integer(), all | sets:set(user_id()), voice_state(), guild_state()
 ) -> ok.
-sync_voice_state_permissions_if_role_matches(GuildId, RoleUsers, UserId, VoiceState, State) ->
-    case sets:is_element(UserId, RoleUsers) of
+maybe_sync_affected_voice_state(GuildId, Affected, VoiceState, State) ->
+    case voice_state_utils:voice_state_user_id(VoiceState) of
+        undefined -> ok;
+        UserId -> sync_if_affected(GuildId, UserId, Affected, VoiceState, State)
+    end.
+
+-spec sync_if_affected(
+    integer(), user_id(), all | sets:set(user_id()), voice_state(), guild_state()
+) -> ok.
+sync_if_affected(GuildId, UserId, Affected, VoiceState, State) ->
+    case is_affected(UserId, Affected) of
         true -> sync_voice_state_permissions(GuildId, UserId, VoiceState, State);
         false -> ok
     end.
 
--spec user_ids_with_role(integer(), guild_state()) -> sets:set(user_id()).
-user_ids_with_role(RoleId, State) ->
-    Data = maps:get(data, State, #{}),
-    RoleIndex = guild_data_index:member_role_index(Data),
-    sets:from_list(maps:keys(maps:get(RoleId, RoleIndex, #{}))).
+-spec is_affected(user_id(), all | sets:set(user_id())) -> boolean().
+is_affected(_UserId, all) -> true;
+is_affected(UserId, Affected) -> sets:is_element(UserId, Affected).
 
 -spec get_member_user_id(map()) -> user_id() | undefined.
 get_member_user_id(MemberUpdate) ->
@@ -377,7 +380,7 @@ maybe_sync_permissions_on_member_update_no_role_change_test() ->
     },
     ?assertEqual(ok, maybe_sync_permissions_on_member_update(MemberUpdate, State)).
 
-maybe_sync_permissions_on_role_update_uses_role_index_test() ->
+sync_roles_after_change_uses_role_index_test() ->
     TestFun = make_sync_test_fun(),
     GuildId = 42,
     RoleId = 999,
@@ -408,9 +411,7 @@ maybe_sync_permissions_on_role_update_uses_role_index_test() ->
             GuildId, RoleId, OtherRoleId, UserId, OtherUserId, ChannelId, Permissions
         )
     },
-    ok = maybe_sync_permissions_on_role_update(
-        #{<<"id">> => integer_to_binary(RoleId)}, State
-    ),
+    ok = sync_roles_after_change([RoleId], State, State),
     receive
         {synced, GuildId, ChannelId, UserId, <<"conn1">>, _Perms} -> ok
     after 200 ->

fluxer_gateway/test/gateway_cluster_membership_tests.erl

@@ -291,6 +291,51 @@ update_members_role_lookups_run_concurrently_test() ->
         ?assertEqual(lists:usort(Nodes), maps:get(sessions, maps:get(members_by_role, State1)))
     end).
 
+member_nodedown_starts_presence_grace_and_nodeup_cancels_it_test() ->
+    with_clean_cluster_terms(fun() ->
+        with_presence_cache(fun(CachePid) ->
+            Peer = 'peer@a',
+            State0 = (base_state(#{members => lists:usort([node(), Peer])}))#{
+                discovered => [Peer]
+            },
+            {noreply, State1} = gateway_cluster_membership:handle_info(
+                {nodedown, Peer}, State0
+            ),
+            ?assertNot(lists:member(Peer, maps:get(members, State1))),
+            ?assert(maps:is_key(Peer, pending_nodedown_cleanups(CachePid))),
+            {noreply, State2} = gateway_cluster_membership:handle_info({nodeup, Peer}, State1),
+            ?assert(lists:member(Peer, maps:get(members, State2))),
+            ?assertEqual(#{}, pending_nodedown_cleanups(CachePid))
+        end)
+    end).
+
+non_member_nodedown_does_not_start_presence_grace_test() ->
+    with_clean_cluster_terms(fun() ->
+        with_presence_cache(fun(CachePid) ->
+            State0 = base_state(#{}),
+            {noreply, _State1} = gateway_cluster_membership:handle_info(
+                {nodedown, 'remsh@host'}, State0
+            ),
+            ?assertEqual(#{}, pending_nodedown_cleanups(CachePid))
+        end)
+    end).
+
+with_presence_cache(Fun) ->
+    {ok, Pid} =
+        case whereis(presence_cache) of
+            undefined -> presence_cache:start_link();
+            Existing -> {ok, Existing}
+        end,
+    try
+        Fun(Pid)
+    after
+        unlink(Pid),
+        gen_server:stop(Pid)
+    end.
+
+pending_nodedown_cleanups(CachePid) ->
+    maps:get(pending_nodedown_cleanups, sys:get_state(CachePid)).
+
 base_state(Overrides) ->
     Defaults = #{
         members => [node()],

fluxer_gateway/test/guild_request_members_filter_tests.erl

@@ -210,6 +210,143 @@ check_guild_request_rate_limit_invalid_guild_test() ->
         guild_request_members_filter:check_guild_request_rate_limit(invalid_guild_id())
     ).
 
+handle_request_consumes_user_budget_test() ->
+    with_rate_limits_enabled(fun() ->
+        UserId = 555000001,
+        clear_request_rate_limit(UserId),
+        SessionState = #{user_id => integer_to_binary(UserId), guilds => #{}, bot => false},
+        Data = #{<<"guild_id">> => <<"777000001">>, <<"query">> => <<"a">>, <<"limit">> => 10},
+        Results = [
+            guild_request_members:handle_request(Data, self(), SessionState)
+         || _ <- lists:seq(1, ?REQUEST_MEMBERS_RATE_LIMIT_MAX_EVENTS)
+        ],
+        ?assertEqual(
+            lists:duplicate(?REQUEST_MEMBERS_RATE_LIMIT_MAX_EVENTS, {error, guild_not_found}),
+            Results
+        ),
+        ?assertEqual(
+            {error, rate_limited},
+            guild_request_members:handle_request(Data, self(), SessionState)
+        ),
+        clear_request_rate_limit(UserId)
+    end).
+
+handle_request_refuses_guild_over_budget_test() ->
+    with_rate_limits_enabled(fun() ->
+        UserId = 555000002,
+        GuildIds = [777000002, 777000003],
+        SessionState = spent_guild_session(UserId, GuildIds),
+        [GuildIdBin | _] = GuildIdBins = [integer_to_binary(Id) || Id <- GuildIds],
+        ?assertEqual(
+            {error, rate_limited},
+            guild_request_members:handle_request(
+                #{<<"guild_id">> => GuildIdBin, <<"query">> => <<"a">>}, self(), SessionState
+            )
+        ),
+        ?assertEqual(
+            {error, guild_not_found},
+            guild_request_members:handle_request(
+                #{<<"guild_ids">> => GuildIdBins, <<"query">> => <<"a">>}, self(), SessionState
+            )
+        ),
+        clear_request_rate_limit(UserId),
+        lists:foreach(fun clear_guild_request_rate_limit/1, GuildIds)
+    end).
+
+spent_guild_session(UserId, GuildIds) ->
+    clear_request_rate_limit(UserId),
+    ensure_ets_table(?REQUEST_MEMBERS_GUILD_RATE_LIMIT_TABLE),
+    Now = erlang:system_time(millisecond),
+    Spent = lists:duplicate(?REQUEST_MEMBERS_GUILD_RATE_LIMIT_MAX_EVENTS, Now - 1000),
+    ets:insert(?REQUEST_MEMBERS_GUILD_RATE_LIMIT_TABLE, [{Id, Spent} || Id <- GuildIds]),
+    #{
+        user_id => integer_to_binary(UserId),
+        guilds => maps:from_list([{Id, {self(), make_ref()}} || Id <- GuildIds]),
+        bot => false
+    }.
+
+handle_request_budgets_disabled_by_env_test() ->
+    OldValue = os:getenv("FLUXER_DISABLE_RATE_LIMITS"),
+    os:putenv("FLUXER_DISABLE_RATE_LIMITS", "true"),
+    try
+        UserId = 555000004,
+        clear_request_rate_limit(UserId),
+        SessionState = #{user_id => integer_to_binary(UserId), guilds => #{}, bot => false},
+        Data = #{<<"guild_id">> => <<"777000004">>, <<"query">> => <<"a">>},
+        Results = [
+            guild_request_members:handle_request(Data, self(), SessionState)
+         || _ <- lists:seq(1, ?REQUEST_MEMBERS_RATE_LIMIT_MAX_EVENTS + 1)
+        ],
+        ?assertEqual(
+            lists:duplicate(
+                ?REQUEST_MEMBERS_RATE_LIMIT_MAX_EVENTS + 1, {error, guild_not_found}
+            ),
+            Results
+        )
+    after
+        restore_env("FLUXER_DISABLE_RATE_LIMITS", OldValue)
+    end.
+
+rate_limit_tables_outlive_the_request_worker_test() ->
+    with_rate_limits_enabled(fun() ->
+        UserId = 555000005,
+        drop_table(?REQUEST_MEMBERS_RATE_LIMIT_TABLE),
+        {ok, OwnerPid} = start_ets_owner(),
+        try
+            ok = check_rate_limit_in_worker(UserId),
+            ?assertMatch(
+                [{UserId, [_]}], ets:lookup(?REQUEST_MEMBERS_RATE_LIMIT_TABLE, UserId)
+            ),
+            ?assertEqual(OwnerPid, ets:info(?REQUEST_MEMBERS_RATE_LIMIT_TABLE, owner))
+        after
+            gen_server:stop(OwnerPid),
+            drop_table(?REQUEST_MEMBERS_RATE_LIMIT_TABLE)
+        end
+    end).
+
+check_rate_limit_in_worker(UserId) ->
+    Self = self(),
+    {Worker, Ref} = spawn_monitor(fun() ->
+        Self ! {checked, guild_request_members_filter:check_request_rate_limit(UserId)}
+    end),
+    receive
+        {checked, ok} -> ok
+    after 1000 -> error(worker_timeout)
+    end,
+    receive
+        {'DOWN', Ref, process, Worker, _} -> ok
+    after 1000 -> error(worker_alive)
+    end.
+
+drop_table(Name) ->
+    case ets:whereis(Name) of
+        undefined -> ok;
+        _ -> ets:delete(Name)
+    end.
+
+prune_expired_entries_drops_idle_keys_test() ->
+    Table = ets:new(request_members_prune_test, [public, set]),
+    Now = erlang:system_time(millisecond),
+    ets:insert(Table, [
+        {idle_window, [Now - 20000, Now - 25000]},
+        {active_window, [Now - 100, Now - 20000]},
+        {idle_bot, Now - 40000},
+        {active_bot, Now - 100}
+    ]),
+    ?assertEqual(ok, guild_request_members_filter:prune_expired_entries(Table, 10000)),
+    ?assertEqual(
+        [active_bot, active_window],
+        lists:sort([Key || {Key, _} <- ets:tab2list(Table)])
+    ),
+    ets:delete(Table),
+    ?assertEqual(gone, guild_request_members_filter:prune_expired_entries(Table, 10000)).
+
+start_ets_owner() ->
+    case whereis(guild_ets_owner) of
+        undefined -> guild_ets_owner:start_link();
+        Existing -> {ok, Existing}
+    end.
+
 restore_env(Key, false) ->
     os:unsetenv(Key);
 restore_env(Key, Value) ->

fluxer_gateway/test/guild_sessions_connect_tests.erl

@@ -141,6 +141,72 @@ normalize_connect_queue_test() ->
     ?assertEqual(Q, guild_sessions_connect_cleanup:normalize_connect_queue(Q)),
     ?assertEqual(undefined, guild_sessions_connect_cleanup:normalize_connect_queue(undefined)).
 
+session_down_clears_passive_sync_entry_test() ->
+    ok = passive_sync_registry:init(),
+    SessionId = <<"passive-down">>,
+    ok = passive_sync_registry:store(SessionId, 42, stored_passive_state()),
+    {Ref, Session, State} = down_state(SessionId),
+    {noreply, _} = guild_sessions_connect:handle_session_down(Ref, {SessionId, Session}, State),
+    ?assertEqual([], ets:lookup(passive_sync_registry, {SessionId, 42})).
+
+remove_session_clears_passive_sync_entry_test() ->
+    ok = passive_sync_registry:init(),
+    SessionId = <<"passive-remove">>,
+    ok = passive_sync_registry:store(SessionId, 42, stored_passive_state()),
+    _ = guild_sessions_connect:remove_session(SessionId, remove_session_state(SessionId)),
+    ?assertEqual([], ets:lookup(passive_sync_registry, {SessionId, 42})).
+
+passive_sync_registry_stays_bounded_across_session_cycles_test() ->
+    ok = passive_sync_registry:init(),
+    GuildRows = fun() -> length(ets:match(passive_sync_registry, {{'_', 42}, '_'})) end,
+    Before = GuildRows(),
+    lists:foreach(
+        fun(N) ->
+            SessionId = <<"cycle-", (integer_to_binary(N))/binary>>,
+            ok = passive_sync_registry:store(SessionId, 42, stored_passive_state()),
+            {Ref, Session, State} = down_state(SessionId),
+            {noreply, _} = guild_sessions_connect:handle_session_down(
+                Ref, {SessionId, Session}, State
+            )
+        end,
+        lists:seq(1, 500)
+    ),
+    ?assertEqual(Before, GuildRows()).
+
+guild_terminate_clears_passive_sync_entries_test() ->
+    ok = passive_sync_registry:init(),
+    GuildId = 4242,
+    OtherGuildId = 4343,
+    lists:foreach(
+        fun({SessionId, GId}) ->
+            ok = passive_sync_registry:store(SessionId, GId, stored_passive_state())
+        end,
+        [{<<"t1">>, GuildId}, {<<"t2">>, GuildId}, {<<"t1">>, OtherGuildId}]
+    ),
+    ok = guild:terminate(shutdown, #{
+        id => GuildId,
+        sessions => #{<<"t1">> => #{}, <<"t2">> => #{}}
+    }),
+    ?assertEqual([], ets:lookup(passive_sync_registry, {<<"t1">>, GuildId})),
+    ?assertEqual([], ets:lookup(passive_sync_registry, {<<"t2">>, GuildId})),
+    ?assertEqual(
+        stored_passive_state(), passive_sync_registry:lookup(<<"t1">>, OtherGuildId)
+    ),
+    ok = passive_sync_registry:delete(<<"t1">>, OtherGuildId).
+
+stored_passive_state() ->
+    #{
+        previous_passive_updates => #{<<"1">> => <<"2">>},
+        previous_passive_channel_versions => #{},
+        previous_passive_voice_states => #{}
+    }.
+
+down_state(SessionId) ->
+    State0 = remove_session_state(SessionId),
+    Session = maps:get(SessionId, maps:get(sessions, State0)),
+    Ref = maps:get(mref, Session),
+    {Ref, Session, State0#{guild_session_refs => #{Ref => SessionId}}}.
+
 remove_session_state(SessionId) ->
     SessionData = #{
         session_id => SessionId,

fluxer_gateway/test/presence_cache_tests.erl

@@ -90,6 +90,14 @@ nodeup_cancels_grace_period_test() ->
     timer:sleep(50),
     ?assertEqual(ok, gen_server:stop(Pid)).
 
+nodedown_grace_expiry_clears_pending_cleanup_test() ->
+    {ok, Pid} = maybe_start_for_test(),
+    ok = presence_cache:handle_nodedown('lost@node'),
+    ?assertMatch(#{'lost@node' := _}, maps:get(pending_nodedown_cleanups, sys:get_state(Pid))),
+    Pid ! {nodedown_grace_expired, 'lost@node'},
+    ?assertEqual(#{}, maps:get(pending_nodedown_cleanups, sys:get_state(Pid))),
+    ?assertEqual(ok, gen_server:stop(Pid)).
+
 anti_entropy_no_op_when_in_sync_test() ->
     {ok, Pid} = maybe_start_for_test(),
     State = cache_state(sys:get_state(Pid)),